Melting a Terminatrix



Melting a Terminatrix
Melting a Terminatrix
Nigel Sumner, Samir Hoon, Willi Geiger, Sebastian Marino, Nick Rasmussen, Ron Fedkiw ∗
Industrial Light + Magic
For Terminator 3: Rise of the Machines, one of the many challenging sequences ILM was presented with depicted the films antagonist, the ”TX”, struggling to free itself from the pull of a magnetic
field. The TX embodies the features of both the previous films assailants, being comprised of both an armored endoskeleton and a
liquid metal exterior. The primary challenge in the sequence was
to digitally create the gradual liquefaction of the TX’s exterior. To
achieve this we turned to a fluid simulation approach, utilizing a
particle level set method [Enright et al. 2002; Foster and Fedkiw
2001]. While this gave us the means to produce the desired dynamic effects, production requirements necessitated the development and implementation of techniques and tools that allow artists
to control every aspect of the simulation. Developing intuitive and
accurate control is essential to achieving an aesthetic vision while
producing a dynamically realistic simulation. This sketch describes
the production processes and techniques developed to complete this
While a variety of techniques were being used to create liquid metal
effects in the film using both particle simulations and animated geometry, none of these techniques provided the detailed structure or
the dynamic nature required to produce a plausible fluid flow and
suspend the audience’s disbelief. To achieve realistic dynamics, we
turned instead to a physics-based 3d fluid simulation engine based
on the particle level set technique. This method uses a level set
stored on a regular volume grid to represent the fluid/air interface,
which is augmented with particles on either side to help prevent
volume loss as the level set is advected through the fluid velocity field. Identifying the level of control needed over the simulation
parameters became an evolving process, the requirements of the engine growing in parallel to the requirements of the shots. A method
for inputting this information into the engine was required, and it
needed to be extensible on both a global and local definition scale.
Simulation and Control
We used Maya particles to control and direct the fluid. A poisson
distribution technique was used to evenly disperse particles over
the outer surfaces of our creature, and these particles carried animation dependent properties such as position, velocity, viscosity
and surface normals. This provided us with a generic method for
inputting control properties into the fluid. We rasterized the geometry in order to describe both the fluid’s initial condition and the
endoskeleton interior as a collision volume. The exported particles
were used as global or local controls for defining regions of the flow
with varying viscosity and areas where the fluid was constrained to
track the particle motion. Geometry was used as fluid erasers or
emitters to enhance the degree of control. To produce simulation
results with a reasonable resolution within a practical turn around
time, we employed various techniques. The engine was not multithreaded, and we instead divided the simulation domain into several
smaller (possibly moving) overlapping domains. To cope with localized simulation over large domains, we implemented grid sourcing. That is, a pre-simmed higher resolution, but smaller domain
∗ e-mail:
[email protected], [email protected]
Figure 1: Various aspects of our pipeline highlighting control, preview, texture and liquid metal.
was used to generate level sets that in turn act as input into a larger
secondary domain thereby increasing the physical simulation area
covered. The overhead of collision grid lookup was reduced significantly by caching out composite grids increasing engine performance and reducing memory overhead.
We render the implicit surface defined by the level set directly by
ray tracing. This is very efficient since the level set defines the distance to the fluid surface at all points in the volume. The fluid is
rendered within the final scene so we can include reflections and
occlusions of other objects. A potential difficulty with rendering
level sets is the lack of texture coordinates. This was overcome by
advecting particles through the fluid, which carry texture coordinates and any other information the artist desires. The information
from the particles nearest each intersection point is blended and
used to define texture coordinates and attributes to control the surface appearance. Because the particles have been advected through
the fluid, the textures flow smoothly with the surface.
E NRIGHT, D., M ARSCHNER , S., AND F EDKIW, R. 2002. Animation and rendering
of complex water surfaces. In Proceedings of SIGGRAPH 2002, 736–744.
F OSTER , N., AND F EDKIW, R. 2001. Practical animation of liquids. In Proceedings
of SIGGRAPH 2001, 15–22.

Similar documents